关于macOS构建问题的openvino

Question

我按照github https://github.com/openvinotoolkit/openvino上的说明在macOS上构建了它，并完成了它，当我写了一小段代码来检查一切是否清晰

#include <string>
#include <iostream>
#include <dlfcn.h>
#include <fstream>

#define USE_STATIC_IE

#define _CRT_SECURE_NO_WARNINGS

#include <ie_core.hpp>

using namespace :: std;

int main(){
    dlopen("/Users/nk/Documents/HZ/openvino/inference-engine/temp/tbb/lib/libtbb.dylib", RTLD_LAZY);
    dlopen("/Users/nk/Documents/HZ/openvino/bin/intel64/Release/lib/libngraph.dylib", RTLD_LAZY);
    dlopen("/Users/nk/Documents/HZ/openvino/bin/intel64/Release/lib/libinference_engine_transformations.dylib", RTLD_LAZY);
    dlopen("/Users/nk/Documents/HZ/openvino/bin/intel64/Release/lib/libinference_engine_legacy.dylib", RTLD_LAZY);
    dlopen("/Users/nk/Documents/HZ/openvino/bin/intel64/Release/lib/libinference_engine.dylib", RTLD_LAZY);
    dlopen("/Users/nk/Documents/HZ/openvino/bin/intel64/Release/lib/libinference_engine_lp_transformations.dylib", RTLD_LAZY);
    dlopen("/Users/nk/Documents/HZ/openvino/bin/intel64/Release/lib/libMKLDNNPlugin.dylib", RTLD_LAZY);

    const std::string pluginsFilePath = "/Users/nk/Documents/HZ/openvino/bin/intel64/Release/lib/plugins.xml";
    InferenceEngine::Core ie(pluginsFilePath);
    const auto info = ie.GetVersions("CPU");

    if(!info.empty())
        cout << "Yes";

    return 0;
}

当我使用命令编译它时

g++ -std=c++11 -I/Users/nk/Documents/HZ/openvino/inference-engine/include testdll.cpp -o testdll && "/Users/nk/Documents/FelenaSoft/"testdll

它给了我这个错误

Undefined symbols for architecture x86_64:
  "InferenceEngine::Core::Core(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&)", referenced from:
      _main in testdll-788073.o
  "InferenceEngine::Core::GetVersions(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&) const", referenced from:
      _main in testdll-788073.o
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)

然后，我添加了.a库并运行以下命令

g++ -std=c++11 -L/Users/nk/Documents/HZ/openvino/bin/intel64/Release/lib -lmkldnn -lpugixml -linference_engine_s -I/Users/nk/Documents/HZ/openvino/inference-engine/include testdll.cpp -o testdll && "/Users/nk/Documents/FelenaSoft/"testdll 

我得到了这个

Undefined symbols for architecture x86_64:
  "ngraph::as_output_vector(std::__1::vector<std::__1::shared_ptr<ngraph::Node>, std::__1::allocator<std::__1::shared_ptr<ngraph::Node> > > const&)", referenced from:
      ngraph::Node::generate_adjoints(ngraph::autodiff::Adjoints&, std::__1::vector<std::__1::shared_ptr<ngraph::Node>, std::__1::allocator<std::__1::shared_ptr<ngraph::Node> > > const&) in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::match_node(ngraph::pattern::Matcher*, ngraph::Output<ngraph::Node> const&)", referenced from:
      vtable for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::match_value(ngraph::pattern::Matcher*, ngraph::Output<ngraph::Node> const&, ngraph::Output<ngraph::Node> const&)", referenced from:
      vtable for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::constant_fold(std::__1::vector<ngraph::Output<ngraph::Node>, std::__1::allocator<ngraph::Output<ngraph::Node> > >&, std::__1::vector<ngraph::Output<ngraph::Node>, std::__1::allocator<ngraph::Output<ngraph::Node> > > const&)", referenced from:
      vtable for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::set_arguments(std::__1::vector<ngraph::Output<ngraph::Node>, std::__1::allocator<ngraph::Output<ngraph::Node> > > const&)", referenced from:
      ExecGraphInfoSerialization::ExecutionNode::clone_with_new_inputs(std::__1::vector<ngraph::Output<ngraph::Node>, std::__1::allocator<ngraph::Output<ngraph::Node> > > const&) const in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::set_output_type(unsigned long, ngraph::element::Type const&, ngraph::PartialShape const&)", referenced from:
      ExecGraphInfoSerialization::ExecutionNode::clone_with_new_inputs(std::__1::vector<ngraph::Output<ngraph::Node>, std::__1::allocator<ngraph::Output<ngraph::Node> > > const&) const in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::m_next_instance_id", referenced from:
      std::__1::shared_ptr<ExecGraphInfoSerialization::ExecutionNode> std::__1::shared_ptr<ExecGraphInfoSerialization::ExecutionNode>::make_shared<>() in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::validate_and_infer_types()", referenced from:
      vtable for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::evaluate(std::__1::vector<std::__1::shared_ptr<ngraph::runtime::HostTensor>, std::__1::allocator<std::__1::shared_ptr<ngraph::runtime::HostTensor> > > const&, std::__1::vector<std::__1::shared_ptr<ngraph::runtime::HostTensor>, std::__1::allocator<std::__1::shared_ptr<ngraph::runtime::HostTensor> > > const&)", referenced from:
      vtable for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::~Node()", referenced from:
      ExecGraphInfoSerialization::ExecutionNode::~ExecutionNode() in libinference_engine_s.a(ie_rtti.cpp.o)
      ExecGraphInfoSerialization::ExecutionNode::~ExecutionNode() in libinference_engine_s.a(ie_rtti.cpp.o)
      std::__1::__shared_ptr_emplace<ExecGraphInfoSerialization::ExecutionNode, std::__1::allocator<ExecGraphInfoSerialization::ExecutionNode> >::~__shared_ptr_emplace() in libinference_engine_s.a(ie_rtti.cpp.o)
      std::__1::__shared_ptr_emplace<ExecGraphInfoSerialization::ExecutionNode, std::__1::allocator<ExecGraphInfoSerialization::ExecutionNode> >::~__shared_ptr_emplace() in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::is_dynamic() const", referenced from:
      vtable for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::description() const", referenced from:
      vtable for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::is_constant() const", referenced from:
      vtable for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::get_arguments() const", referenced from:
      vtable for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::get_output_size() const", referenced from:
      ExecGraphInfoSerialization::ExecutionNode::clone_with_new_inputs(std::__1::vector<ngraph::Output<ngraph::Node>, std::__1::allocator<ngraph::Output<ngraph::Node> > > const&) const in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::write_description(std::__1::basic_ostream<char, std::__1::char_traits<char> >&, unsigned int) const", referenced from:
      vtable for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::copy_with_new_args(std::__1::vector<std::__1::shared_ptr<ngraph::Node>, std::__1::allocator<std::__1::shared_ptr<ngraph::Node> > > const&) const", referenced from:
      vtable for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::get_output_element_type(unsigned long) const", referenced from:
      ExecGraphInfoSerialization::ExecutionNode::clone_with_new_inputs(std::__1::vector<ngraph::Output<ngraph::Node>, std::__1::allocator<ngraph::Output<ngraph::Node> > > const&) const in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::get_default_output_index() const", referenced from:
      vtable for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::get_output_partial_shape(unsigned long) const", referenced from:
      ExecGraphInfoSerialization::ExecutionNode::clone_with_new_inputs(std::__1::vector<ngraph::Output<ngraph::Node>, std::__1::allocator<ngraph::Output<ngraph::Node> > > const&) const in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::get_autob() const", referenced from:
      vtable for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "ngraph::Node::is_output() const", referenced from:
      vtable for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "typeinfo for ngraph::Node", referenced from:
      typeinfo for ExecGraphInfoSerialization::ExecutionNode in libinference_engine_s.a(ie_rtti.cpp.o)
  "vtable for ngraph::Node", referenced from:
      std::__1::shared_ptr<ExecGraphInfoSerialization::ExecutionNode> std::__1::shared_ptr<ExecGraphInfoSerialization::ExecutionNode>::make_shared<>() in libinference_engine_s.a(ie_rtti.cpp.o)
  NOTE: a missing vtable usually means the first non-inline virtual member function has no definition.
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)

有没有人和我有同样的问题，有没有人知道怎么解决？

Answer 1

首先，也是最重要的，我需要对您的工具包安装进行交叉检查。既然您已经在尝试运行一个模型，我假设您已经正确地设置了它(仅供参考的最新版本是2020.4)：https://docs.openvinotoolkit.org/latest/openvino_docs_install_guides_installing_openvino_macos.html

你得到的错误，架构x86_64的未定义符号是Xcode构建失败，表明你在推理引擎中遗漏了一些东西，或者你将模型与某些资源链接的方式是不正确的，或者是其他一些原因。只要这个错误仍然存在，你就不能编译你的代码。这就是为什么链接器命令失败并返回退出代码1的原因。

大多数错误的发生是因为用户没有正确设置OpenVINO环境变量或配置模型优化器。这就是为什么运行验证脚本来验证安装和编译示例非常重要的原因。

在运行任何自定义模型之前，最好有一个可靠的Openvino环境，因此请帮助检查您是否可以像这里的示例https://docs.openvinotoolkit.org/latest/openvino_docs_install_guides_installing_openvino_macos.html那样运行默认模型

默认情况下，你应该已经有了预先训练好的模型，包括开放动物园模型，模型优化器和推理引擎，这些都是的工作示例，如果你正确地配置了工具包，就可以使用。它们应该已经位于您安装的工具包的目录中，您不必将其与其他资源分开安装。

请注意您需要: CMake 3.4或更高版本，Python3.5或更高版本，Apple Xcode命令行工具*，(可选) Apple Xcode* IDE ( OpenVINO不需要，但对开发很有用)，

支持的MAC OS为macOS 10.14.4*

同样，请帮助交叉检查您在Openvino环境中拥有的内容以及使用官方文档https://docs.openvinotoolkit.org/latest/openvino_docs_install_guides_installing_openvino_macos.html进行配置的步骤

谢谢!